The BnaZNRIL population, consisting of 184 F7 lines, was derived by single-seed descent from a cross between two sequenced rapeseed cultivars, Zhongshuang11 (de novo sequencing) and No. 73290 (resequencing) (Yang et al., 2016). As 9 lines were removed from this population for the lack of seeds, 175 lines plus the two parents were used in this study.

Plump and uniform rapeseed seeds were sowed on medical gauze that was fixed to a blue plastic basin (60 × 40 × 15 cm, length × width × height) filled with quarter-strength modified Hoagland's solution (Hoagland and Arnon, 1950). The modified Hoagland's solution (the concentration of N was 15 mM) consisted of: 5 mM Ca(NO₃)₂·4H₂O, 5 mM KNO₃, 2 mM MgSO₄·7H₂O, 1 mM KH₂PO₄, 0.05 mM EDTA-Fe, 46 μM H₃BO₃, 9.14 μM MnCl₂·4H₂O, 0.77 μM ZnSO₄·7H₂O, 0.37 μM NaMoO₄·2H₂O, and 0.32 μM CuSO₄·5H₂O. Six days after sowing as described by Dun et al. (2016), uniform seedlings were selected and transplanted into smaller blue plastic basins (34 × 26 × 12 cm, length × width × height) containing quarter-strength nutrient solution (two N treatments, HN and LN) under the natural condition with a removable rain-shelter. Each basin contained 24 seedlings of 4 lines (six seedlings for each line). As a total, 90 basins were used in each independent experiment. Nutrient solution was renewed once a week. For HN treatment, the quarter-strength and half-strength nutrient solution was used at the first 2 weeks respectively, and full-strength nutrient solution was used until harvest. For LN treatment, the concentration of N was adjusted to 0.5 mM by reducing KNO₃ and replacing Ca(NO₃)₂ by CaCl₂, while K⁺ was complemented by adding K₂SO₄ (refer to Stahl, 2015). The pH value was adjusted to 5.8 ± 0.2 with NaOH or HCl.

Three independent hydroponic culture experiments with a completely random design were carried out at Oil Crops Research Institute of the Chinese Academy of Agricultural Sciences, Wuhan, PR China. The plants were harvested with five fully expanding leaves under HN condition (~37 days after sowing). Accordingly, the plants under LN condition showed typical N-deficiency symptoms. The first experiment (E1) was carried out during the period from October 18 to November 23 in 2015 (37 days), the second (E2) from March 7 to April 14 in 2016 (39 days), and the third (E3) from October 2 to November 4 in 2016 (34 days). During the three experiments, the average temperatures and the average humidities in Wuhan were 19/11°C, 19/10°C, 21/15°C (day/night), and 79, 78, 83%, respectively. Besides, the total sunshine times in Wuhan were 158, 187, and 207 h, respectively, among the three experiments and the average light intensities were about 600–800 μmol•m⁻²•s⁻¹.

At harvest, four uniform plants of each line were removed from the basin. After they had been sampled, the total roots were separated from the shoot base and primary root length (PRL) was investigated manually using a ruler, while shoots were over dried at 80°C until a constant weight to evaluate shoot dry weight (SDW). The intact root system were immersed and dispersed in a transparent plastic tray with water for scanning with a scanner (EPSON V700, Japan), and total root length (TRL), total root surface area (TSA), total root volume (TRV), total root number (TRN) were analyzed using WinRHIZO software (Pro, 2012b, Canada). Finally, roots were over dried at 80°C to evaluate root dry weight (RDW). Besides, total dry weight (TDW) and root-shoot ratio in dry weight (RSRD) were calculated.

The dried root and shoot samples were separately ground into powder, about 0.1 g were weighted and dissolved in H₂SO₄-H₂O₂, and then diluted with pure water to 1.25 L. Subsequently, N concentration (mg/L) were analyzed using Smartchem 200 automatic analyzer (Westco Scientific Instruments, Westco). Each sample was measured with three repetitions. RNC or SNC was N content (mg) in per unit weight (g) of root or shoot sample respectively and was calculated as follows: RNC or SNC = (sample N concentration × 1.25) / sample weight. The N uptake of root (RNU) and shoot (SNU) were calculated by multiplying weight by N concentration, respectively. Total N uptake (TNU) or N utilization efficiency (NUtE) was obtained by adding RNU and SNU together or by dividing TDW by TNU. All the traits investigated in this study were summarized in Table 1.

A total of 15 phenotypic traits, including 7 RM- and 8 NUE-related traits (Table 1) represented by the means of four plants for each genotype, were used for phenotypic analysis and QTL analysis. The broad-sense heritability (h²) of each measured trait was calculated as h² = σg2/(σg2 + σge2/n + σe2/nr), where σg2, σge2, and σe2 are the variance of genotype, genotype × environment, and error, respectively, and n and r are the number of independent experiments and replications, respectively. The estimation of σg2, σge2, and σe2 were obtained with the software SAS 9.2 (SAS Institute Inc., NC, USA) using the GLM procedure. Correlation analysis and principal component analysis (PCA) were calculated with software SAS 9.2 (SAS Institute Inc., NC, USA) using the PROC CORR and PROC PRINCOMP procedure, respectively.

The linkage mapping of QTLs was performed by composite interval mapping program (Zeng, 1994) using the Windows QTL Cartographer version 2.5 software (http://statgen.ncsu.edu/qtlcart/WQTLCart.htm). The corresponding genetic map consisted of 2264 unique loci/bins, which covered a total length of 2,107 cM distributed on 19 linkage groups (Yang et al., 2016). In this study, a walk speed of 1 cM, 5 control markers, a window size of 10 cM and forward regression method were used. The LOD threshold was determined by permutation analysis with 1,000 repetitions (Churchill and Doerge, 1994). The LOD threshold was set at 3.2–6.2 (p = 0.05) to identify significant QTLs. To avoid missing QTLs with a relatively small effect, a lower LOD threshold was set at 1.7–2.2 (p = 0.50) to detect suggestive QTLs. Both significant QTLs and overlapping suggestive QTLs were admitted (Long et al., 2007) and named as “identified QTL” (Shi et al., 2009).

In this study, “stable QTL (sQTL)” was defined, which was repeatedly detected with more than half of 2-LOD confidence interval overlapping at least two environments or different N levels and has the same additive-effect direction. In addition, sQTLs detected under both HN and LN conditions were called as “constitutive sQTL,” while sQTLs that were detected in at least two environments under either HN or LN conditions, respectively were named as “HN-specific sQTL” or “LN-specific sQTLs,” respectively (Li et al., 2015). These sQTLs were divided into two types: QTLs detected at least once with phenotypic variation explained (R²) ≥ 20% or at least twice with R² ≥ 10% were named as “major sQTL”, and the remainder were named as “minor sQTL” (Price, 2006; Maccaferri et al., 2008; Shi et al., 2009). And QTL meta-analysis was performed using BioMercator 4.2 software (Arcade et al., 2004) to estimate the coincidences of several QTLs for different traits (at least two different traits), which were integrated into the “QTL cluster.” For a QTL cluster, the coincidence of QTLs for two or more traits with same additive-effect directions was considered to be positive, while the coincidence of QTLs with opposite additive-effect directions was considered negative (Coque et al., 2008).

Each identified QTL or QTL cluster was named as “q” + “the name of the trait abbreviation” or “qc” + “the linkage group,” respectively. Arabic numerals were added, if more than one QTL cluster was located on the same linkage group.

Three independent hydroponic experiments for the parental lines and the RIL population were performed to evaluate 7 RM- and 8 NUE-related traits (Table 1) under both HN and LN growth conditions. Although the two parents (Zhongshuang11 and No.73290) showed no significant difference for most of RM- and NUE-related traits under HN level (except in E2, Zhongshuang11 showed significant higher values than No. 73290 in RDW, TRL, TSA, TRV, RNU, SNU, and TNU), they displayed significant differences for PRL, TRL, TSA, TRV, TRN, RDW, and NUtE in all three environments or a single environment under LN level (Table S1).

Continuous phenotypic distribution values with obvious kurtosis among these lines suggesting a quantitative inheritance pattern suitable for QTL identification. Minimum, maximum, mean values and coefficient of variations (CVs) for all investigated traits from each experiment were listed in Table 2. Most traits had considerable phenotypic variation within the BnaZNRIL population as suggested by the CVs ranging from 7.6 to 33.8% (Table 2). Increased PRL (1.6–89.9%), RSRD (54.5–172.7%), TRL (5.2–14.8%), and NUtE (11.5–200%) were response to LN stress, compared with those under HN levels (Table 2), indicating that LN stress stimulated root growth for more N nutrition uptake and improved N utilization, though shoot growth was obviously inhibited. The heritability (h²) of RM-related traits was relatively high, ranging from 0.38 to 0.67 under HN and from 0.48 to 0.60 under LN condition (Table 2). However, the heritability (h²) of NUE-related traits was moderate except SNC and NUtE (low under HN condition, 0.07 and 0.06, respectively), ranging from 0.23 to 0.48 under HN and from 0.26 to 0.59 under LN condition (Table 2). This indicated that RM-related traits are less affected than NUE-related traits on changeable environment under both HN and LN conditions. Although significant genotype × environment was observed, the relatively high heritability (h²) in the majority of the investigated traits indicated the genetic stability of these traits in a genotype among the three repetitions.

Further, for each trait, part of RIL lines had values outside the range between parental lines, suggesting the presence of alleles with positive effects for root development and N utilization in both parents. For individual RM- or NUE-related traits, several of the transgressive lines always displayed extreme values lower or higher than parental lines over the three experiments (for example, B031 had 7–32% lower TRL and RDW than No. 73290, while B152 had 11–41% higher TRL and RDW than ZS11 under LN conditions), providing useful lines for RM or NUE genetic improvement and corresponding molecular mechanism dissection.

The phenotypic correlations among all examined traits were calculated and the principal component analyses (PCA) for all investigated traits were performed to reveal the genetic correlation among RM- and NUE-related traits. For RM-related traits, the majority of traits had strong and significantly correlations (r = 0.71–0.93 under HN levels and r = 0.61–0.95 under LN conditions, P < 0.0001) with each other irrespective of N levels with the exception of PRL and RSRD (Table 3), indicating developmental relevance among these RM-related traits. PRL and RSRD showed smaller or no significant correlation with other RM-related traits under HN condition (r = 0.01–0.44 for PRL, r = 0.02–0.13 for RSRD with others). However, PRL and RSRD had significantly correlations with other RM-related traits under LN condition (r = 0.17–0.54 for PRL, P < 0.05; r = 0.25–0.38 for RSRD, P < 0.05; except r between RSRD and TRN) (Table 3), revealing the change of root system response to N stress, including PRL increasing and faster root development. For NUE-related traits, the dry biomass traits (DB, including SDW and TDW) had high values of correlation coefficients with N uptake traits (NU), SNU (r = 0.82–0.95), RNU (r = 0.61–0.71), and TNU (r = 0.77–0.95) irrespective of N levels, as expected (Table 3), suggesting that plant growth depend heavily on N uptake. On other hand, DB had no significant correlation with NUtE under HN level, while positive correlation with NUtE (r = 0.28) under LN level, indicating that plants rely on the higher N uptake and NUtE to maintain larger biomass under LN level (Table 3).

Pearson's correlations were also calculated comparing the RM- and NUE-related traits. It was obviously that RM-related traits (except for PRL and RSRD) was significant correlated with NU and DB irrespective of N levels (r = 0.51–0.86 under HN condition, r = 0.43–0.82 under LN condition) (Table 3), indicating the phenotypic relationship between root system, plant growth and N use efficiency. Interestingly, although SNC and RNC showed no significant correlation with RM-related traits under HN condition, they are significantly correlated with several RM-related traits (TRL, TSA, and TRV) under LN condition.

Principle component analysis (PCA) also showed that RM-related traits except PRL and RSRD were all closely related and combined into two groups distributed in two different N levels, respectively (Figure 1). In NUE-related traits, only DB and NU were both assigned into the two RM-related groups. Thus, all these results showed that RM-related traits (except for PRL and RSRD) were more likely to be genetically associated with DB and NU, but not with NUtE, indicating that RM-related traits may play an extremely important role in plant N uptake rather than in plant N utilization.

Using the WinQTL cartographer software, a total of 129 identified QTLs (104 significant QTLs and 25 overlapping suggestive QTLs) associated with the investigated traits was detected on three environments, respectively. Among these, 91 QTLs were assumed to influence RM-related traits and 38 QTLs were assumed to influence NUE-related traits (Figure 2, Tables S2, S3). These QTLs were located across 12 of the 19 chromosomes in B. napus (including A01-A07, A09, C04, C05, C08, and C09). Of these, 45 (~35%) were located on the A07 linkage. More QTLs for NUE-related traits was detected under LN condition than that under HN level (7 under HN condition, 31 under LN condition), while similar number of QTLs for RM-related traits was detected under two N levels (43 under HN condition, 48 under LN condition; Table S3). Consequently, 56 out of 129 identified QTLs (~43% of the total number QTLs) were integrated into 23 sQTLs, which were repeatedly detected at least two environments or different N levels (Figure 2 and Table 4). Among these, 20 sQTLs were associated with RM-related traits and 3 sQTLs were associated with NUE-related traits. Notably, 16 out of 23 sQTLs were considered as “constitutive sQTL,” because they were detected under both HN and LN conditions. These results suggested that QTLs for RM-related traits are less affected than NUE-related traits on changeable environment under both HN and LN conditions.

Most pairs of RM-and NUE-related traits showed significantly genetic correlation in Table 3, providing the potential to identify QTLs with pleiotropy or specificity for these traits. QTL meta-analysis was performed to find pleiotropic “QTL cluster.” As a result, a total of 83 identified QTLs (~64% of the total number) were integrated into 22 distinct pleiotropic QTL clusters (Figure 3, Table 5). The 22 distinct QTL clusters were distributed on eight chromosomes (A01, A02, A04-A07, A09, and C04) with a range of 2–13 identified QTLs. About 68% of these were distributed on chromosome A01 (four), A02 (three), and A07 (eight). Besides, QTL clusters detected for both RM-related traits and NUE-related traits were defined as “RM-NUE QTL cluster,” while QTL clusters that were exclusively detected for RM-related traits or NUE-related traits were named as “RM-specific QTL cluster” or “NUE-specific QTL cluster,” respectively. Thus, all of the 22 distinct QTL clusters were divided into 10 (~45%) RM-specific QTL clusters, 6 (~27%) NUE-specific QTL clusters, and 6 (~27%) RM-NUE QTL clusters.

Root system is the key place for N acquisition (Lynch, 1995). Illuminating the genetic relationship between RM and NUE traits could provide the basis of RM-based approach for NUE genetic improvement. Our study uncovered the tight connection between RM and NUE traits from two aspects. On the one hand, phenotypic correlation analysis and principal component analysis showed high genetic correlation between RM and NUE traits (Figure 1, Table 3). Generally, RM-related traits (except for PRL and RSRD) had high positive correlations with DB and NU irrespective of N levels, but no correlation or little correlation with NUtE under HN or LN level respectively. PCA also showed that RM-related traits (except for PRL and RSRD) and NUE-related traits (except for NC and NUtE) were assigned into the same group under HN and LN conditions, respectively (Figure 1, group 1 and group 2). These results indicated that RM, especially “root size,” play a critical role in plant N uptake, rather than in N utilization, which was consistent with the previous report on maize (Li et al., 2015). On the other hand, the co-localization of QTLs for RM- and NUE-related traits in this study further supports the tight connection between RM and NUE traits. Among the six RM-NUE QTL clusters, four overlapped between RM and NU, two between RM and DB and others between RM and NC, but no between RM and NUtE, and five of them were positive in which all the QTLs had same additive-effect directions (Figure 3, Table 5). These proved that N acquisition, rather than N utilization, is more likely correlated with root morphology.

In conclusion, the significantly phenotypic correlation and the co-localization of QTLs fully demonstrated the presence of credible genetic relationship between RM and NUE in rapeseed. However, the process of plant N utilization is complex, mainly including N uptake by roots and translocation, assimilation and remobilization inside the plant (Xu et al., 2012). RM and root uptake activity (including root vigor, root N transport, etc.) are the major determinants of N acquisition (Glass, 2003; Garnett et al., 2009; Xu et al., 2012), while the N translocation, assimilation and remobilization refer to N metabolism process (Xu et al., 2012). So, given the presence of six NUE-specific QTL clusters in this study (Figure 3, Table 5) and the credible genetic relationship between RM and NUE, our results provided evidences for two NUE improvement approaches in rapeseed: RM-based and N utilization-based direct approaches.

Both RM and NUE are complex traits, susceptible to growth environment, and difficult in directly measuring that restrict genetic improvement of NUE and RM in crops (Hochholdinger and Tuberosa, 2009). Evaluating these traits using the genomic approach and marker-assisted selection are worldwide challenges for NUE improvement (Coque et al., 2008; Liu et al., 2008; Li et al., 2015). In this study, 129 identified QTLs (including 23 stable QTLs and 4 major QTLs) were detected and 83 QTLs of them were integrated into 22 pleiotropic QTL clusters (Figures 2, 3, Tables 4, 5, Tables S2, S3). Five RM-NUE (qcA02-1, qcA07-6 to qcA07-8 and qcA09-1), ten RM-specific (qcA01-1, qcA01-2, qcA02-3, qcA07-1 to qcA07-5, and qcA09-2) and three NUE-specific (qcA01-3, qcA04, and qcA06) QTL clusters with same directions of additive-effect provided the potential to identify pleiotropic genes or different closely-linked genes controlling root development and N uptake. More importantly, the favorable alleles among these QTL clusters can be used directly for genetic improvement of NUE via a marker-assisted selection approach in rapeseed.

The three largest RM-NUE QTL clusters (qcA07-6, qcA07-7, and qcA07-8) were closely distributed on 90.12–107.84 cM of chromosome A07 and all had the favorable alleles coming from No. 73290. These QTL clusters co-localized with two constitutive stable QTLs, sqTRV.A07-2, and sqRDW.A07-2, suggesting that this region may play a key role in root growth and plant N uptake. Besides, three single-nucleotide polymorphisms (SNPs), significantly associated with RSRD, RDW and/or PRL, which were detected under low phosphorus (P) condition (Wang et al., 2017) were located in the similar genomic region with the three QTL clusters, respectively. In addition, Wang et al. (2017) identified another significantly associated SNP for root length, which was most likely co-localized with qcA02-1. Another QTL cluster, qcA09-1, located on 28.96–38.45 cM, was associated with PRL and RNU/TNU, whereas a QTL for PRL have previously been identified nearby (Zhang et al., 2016). Further, a significantly associated SNP for shoot concentration of Na discovered by Bus et al. (2014) was also localized in this genomic region. Consequently, we proposed that these positive RM-NUE QTL clusters were likely to be the common regions for regulating rapeseed RM traits across different genetic backgrounds, and further involved in plant N, P, and Na uptake.

Five RM-specific QTL clusters on chromosome A07 (from qcA07-1 to qcA07-5) were distributed on 62.46–74.22 cM. Seven sQTLs (including three major sQTLs) for RM-related traits were co-localized with these QTL clusters, suggesting the presence of major genes regulating RM in this region. Besides, under low P condition, Wang et al. (2017) identified six significantly associated SNPs with several root traits and Zhang et al. (2016) detected two QTLs for PRL and SDW that were located in the similar genomic region. Interestingly, all the QTLs distributed on 62.46–74.22 cM of chromosome A07 had the favorable alleles coming from the parental line ZS11, while on 90.12–107.84 cM from the parental line No. 73290, indicating that there are two distinct but important regulatory mechanisms for root development on chromosome A07 under this mapping population. On chromosome A01, qcA01-1 and qcA01-2 were distributed on 43.05–50.41 cM and the major stable QTL sqTRN.A01 was co-localized with these QTL clusters, suggesting the presence of major genes regulating RM in this region. The QTL uq.A1_1 for TRL detected by Zhang et al. (2016) using another genetic population was located in the similar genomic region. Thus, given the credible genetic relationship between RM and NUE, these important QTL clusters would have a considerable breeding value for RM-based NUE improvement in rapeseed, though no co-localization with NUE detected in this study.

Our study suggested two different approaches for NUE improvement via marker-assisted selection in rapeseed: RM-based and N utilization-based direct approaches. Previous reports showed that RM-related traits investigated at an early development stage had significant links with NUE, yield and yield components (Li et al., 2015; Xie et al., 2017). These suggested that breeding varieties with large “root size” at seedling stage may be a promising approach to optimize N uptake efficiency in crops. RM-based NUE improvement provided a novel and exciting opportunity for the manipulation of RM via marker-assisted selection QTLs for improving NUE in rapeseed, though direct NUE improvement had made some achievements as the major approach for NUE breeding previously (Xu et al., 2012; Han et al., 2015). The credible genetic relationship between RM and NUE in this study laid a theoretical foundation for RM-based approach on NUE genetic improvement in rapeseed. Furthermore, in this study, ~2.4 times more QTLs were identified for RM-related traits (91) than that for NUE-related traits (38) and all of the four major QTLs and most of stable QTLs (20 out of 23) were detected to be related to RM traits under HN and/or LN levels, suggested that QTLs for RM-related traits are less affected than NUE-related traits on changeable environment under both HN and LN conditions. Thus, regulating RM to improve NUE via marker-assisted selection would be more feasible and reliable than regulating nitrogen efficiency directly in rapeseed. On the other hand, many co-localized QTLs associated with RM and NUE traits have been reported in crops (An et al., 2006; Coque et al., 2008; Liu et al., 2008; Li et al., 2015) and manipulation of RM via marker-assisted selection to improve nutrient efficiency has been successfully used in rice (Wissuwa, 2005; Pariasca-Tanaka et al., 2014; all for phosphorus use efficiency, PUE) and maize (Li et al., 2015, for NUE; Gu et al., 2016, for PUE). Moreover, Li et al. (2015) identified 53 advanced backcross-derived lines (ABLs) containing RM-NUE QTL clusters via marker-assisted selection of RM and the GY/NUE of these ABLs showed apparent mean increases of 13.8% under HN and of 15.9% under LN conditions, compared to recurrent background Wu312. All this indicated that manipulation of RM via marker-assisted selection to improve NUE would be more effective than marker-assisted selection for NUE directly in rapeseed.

Though nutrient absorption and root development of plants are different in hydroponic culture and field experiment (Chapman et al., 2012; Watt et al., 2013; Petrarulo et al., 2015), some of the experimental evidence available has shown their correlations (Lynch and Brown, 2001; Cui et al., 2008). Moreover, QTLs for root traits detected in hydroponics have been successfully used in nutrient efficiency improvement in field (Li et al., 2015, for NUE; Gu et al., 2016, for PUE). We deduced that the QTLs identified in this study could be used for marker-assisted selection in field, and corresponding examination will be conducted in the field in progress.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.